Research a Person

What It Does

Research a Person turns a name into a compiled, cited dossier — an era-by-era life story, a "Who They Are" synthesis, and (for confirmed public figures) a Social Media Presence section that quotes real posts verbatim with their real engagement numbers, not a vague characterization of "an active online presence."

As of W1.44, reach it from the ✦ Deep button on the chat composer, under Gather (it goes out and brings sources back, the same job as Auto-Research with a person as the subject — it does not sit next to Find the Truth, which checks material you already have, under Interrogate). Two stages, always in this order:

1. Free identity confirmation — before anything is charged, you confirm which real person you mean.
2. A paid, footprint-gated research sweep — a background run (roughly 10–40 minutes depending on depth) that gathers from clean public-record sources always, and from consented social platforms only when the confirmed identity has a real public footprint.

The setup always opens fresh. Unlike Auto-Research, Auto-Write, and Detect Conflicts, Research a Person doesn't preserve a half-filled draft if you leave for the chat or a source and come back — leaving and returning starts Stage 1 over. A run already in progress isn't affected (it keeps going and you're taken back to it), and any dossier you've already finished stays reachable in the Reports tab either way.

The result is a person_report artefact that opens in the same reading pane as every other Cafe report, with the one merged action row below the body (as of W1.44) — Copy / Download (.md) / Download PDF / Save to Notebooks / Share / Open full / Add to Tray / Delete, same as any other report.

---

Stage 1 — Confirming Who You Mean

Click ✦ Deep → Research a Person. The setup's first step asks for:

• Name (required)
• Identifying details (optional free text — helps disambiguate a common name, or is the only research material CoffeeScribe has for a non-public person)
• Seed links (optional, up to 5 URLs — a homepage, a Wikipedia page, a social profile)

Click Next and CoffeeScribe runs a free lookup against Wikidata and Wikipedia — no tokens charged, no run created yet. Matching candidates appear as rows (name, short description, dates, occupation) usually in under a couple of seconds.

• Pick a candidate to anchor the run on that specific, verified person.
• Click "see more" on a candidate to expand it and double-check before committing — useful for a common name.
• "None of these — use my details only" — if nobody matches (a private individual, an obscure figure, or a name Wikidata doesn't carry), the run continues on your typed details and any seed links instead of a confirmed identity. This is the seed-link fallback: CoffeeScribe never invents a match, and never silently guesses at a different real person with a similar name.

An ambiguous name (e.g. "John Smith") returns multiple distinct candidates so you can pick the right one — the picker also filters out non-person Wikipedia hits (election pages, disambiguation lists) so you're only ever choosing between actual people.

---

Stage 2 — Confirm: Accounts, Lanes, Focus, Depth

Once you've picked a candidate (or gone details-only), the confirm step shows everything that will run, and its cost, before you spend anything.

Verified accounts card

For a confirmed public figure, CoffeeScribe re-derives their verified social handles server-side (X/Twitter, Instagram, TikTok, YouTube, Facebook, LinkedIn) — this doesn't depend on what loaded in your browser during Stage 1, so a slow or failed lookup earlier can't silently produce a report with zero social coverage. You can also add your own handles here if you know an account CoffeeScribe's lookup missed.

Lane checkboxes

CoffeeScribe plans a set of research lanes based on who the person is — a politician gets a voting-record/funding lane, a musician gets a works-and-reception lane, a deceased figure gets a legacy/estate lane instead of live-social lanes, and so on. The confirm step shows the lanes it's planned to run as tickable checkboxes — life story, social media, own words (interviews/speeches), public perception, and any persona-specific lanes.

You can only narrow this list, never widen it — if the plan didn't offer a social lane (because the person isn't a confirmed public figure), there's no checkbox to force it on. Untick a lane you don't want and the live cost estimate drops immediately.

Custom focus (up to 2)

Below the lane checkboxes, an "Anything specific to focus on?" input lets you add up to two custom research angles in your own words — for example, for a historical figure: "His death" or "His marriage." Each focus becomes its own research lane and its own section in the final report (e.g. "Focus: His death").

A focus is always treated as a topic, never as an instruction to the AI — a focus that asks for contact details, an address, or someone's daily schedule/whereabouts (e.g. "where he lives") is declined with a polite message rather than run. This is enforced the same way everywhere else in CoffeeScribe research: no contact-information or address compilation, full stop.

Depth

Three levels, same outcome-named pattern as Find the Truth and Auto-Research:

Depth changes how many lanes fire and how many queries/results each lane pulls — not just a token ceiling. A live cost estimate (tokens + Apify cents) recalculates as you toggle lanes, add/remove focuses, or change depth.

Responsible-use copy (shown on every run)

Before Start, every run shows this notice:

"People research compiles publicly available information. Use it for legitimate research on public figures. CoffeeScribe never collects contact details or addresses, and limits research on people without a public profile to basic public sources. Don't use this to monitor, harass, or profile private individuals."

Consent

Account-level first (W1.41). If you haven't yet accepted CoffeeScribe's Social Sources Terms for your account, starting a run that could touch social platforms shows that screen before anything else — Accept enables social lanes and records your acceptance, "Not now" runs the research anyway with social lanes left out (named in the run), and dismissing the screen backs out of starting the run entirely. This applies once per account, not per Cafe — see Research Mode → the account-level switch for the full explanation and how to change your choice in Settings.

Then, per-Cafe. If this is the first run in this Cafe to touch a social platform you haven't consented to yet (e.g. Facebook), a one-time consent modal appears — the same per-Cafe, per-platform consent used everywhere else in Research Mode. Once granted, it isn't asked again for that platform in this Cafe. The per-Cafe consent only matters once your account-level switch is on — with it off, there is nothing for the per-Cafe consent to enable.

---

The Run

Click Start and the sweep runs as a background job — close the tab, refresh, or navigate away and it keeps going; reopen the Cafe and progress reattaches automatically. A full sweep typically takes 10–40 minutes depending on depth and how many lanes/platforms are in play; a deep run across every lane on a heavily-documented public figure can take 30–60 minutes.

A progress panel shows the current phase (reading your links → gathering public records → listening to interviews → reading public discussion → sweeping social platforms → confirming the right person → writing the report) — the active phase pulses so the panel visibly reads as alive during a multi-minute run. The panel also sets the expectation up front ("Deep research — this usually takes 5–15 minutes") and tells you it's safe to close the window — click Hide in the footer and the run keeps going in the background; come back to the Cafe any time to check on it. Click Cancel instead if you want to stop the run — it stops at the next safe boundary, sources gathered so far are kept, and no report is produced for a cancelled run.

If a run genuinely stalls or the server restarts mid-run, an automatic housekeeping check marks a run "failed" only when it has gone silent — a slow-but-live deep run is never mistaken for a dead one, because CoffeeScribe tracks real progress, not just elapsed time.

If the system is busy, your run may wait for a free slot first. CoffeeScribe caps how many deep background runs — Research a Person, Find the Truth, and Auto-Research — can be executing at once, platform-wide, so an overloaded moment degrades to waiting rather than to everything slowing down. If the cap is full when you click Start, a "Waiting for a free slot" panel appears instead of the usual progress panel, showing your position in line. It starts automatically once a slot frees up, nothing is charged while you wait, you can close the window and it keeps its place, and a Cancel button lets you give up your spot. This is uncommon — you'll only see it during a burst of simultaneous deep runs.

The sources rail and reading pane refresh automatically when the run settles (W1.17). A couple of seconds after the run finishes (report ready, error, or stopped), the Cafe's sources rail and reading pane refresh themselves — you don't need to manually reload the page to see the run's gathered sources or open its report.

---

The Report

When the run finishes, the person_report opens in the reading pane. It's built dynamically from what the run actually found — sections only appear when there's real evidence behind them:

• Era-by-era life story — Overview, Early Life & Background, Life & Career, Legacy / Current Standing.
• Social Media Presence (confirmed public figures with consented platforms only) — per-platform stats, and the actual text of the most-liked and most notable/controversial posts, quoted verbatim, with their real engagement numbers (e.g. "his most-liked post — '…' (1.27M likes)"), each footnoted.
• Who They Are — a synthesis characterization, grounded only in what was actually gathered.
• Persona sections — whichever apply: Works & Reception (creators), Wealth & Business, Political Views & Values, Death & Aftermath (deceased subjects — see below), Family, Controversies & Legal Issues, and any Focus: <your text> section for a custom focus you added.
• References — a numbered bibliography, listed in a collapsible disclosure at the bottom of the report, collapsed by default (click to expand). This applies to every research report type, not just person reports — it keeps a long dossier's reference list from dominating the page. In-text [N] citation clicks still open the right entry even while it's collapsed.

Guardrails you'll see reflected in the report

• No contact info, ever. Email, phone, and address-shaped data are stripped before anything is stored, and the writing prompt is forbidden from including a contact section — this is enforced at two separate points (ingest and write), not just one.
• Public-figure gating. Social compilation only happens for a footprint CoffeeScribe re-confirms server-side. A private or unconfirmed person's run stays clean-core-only (public web/news/Wikipedia sources) and the report opens with an honest limited-coverage note explaining why.
• Unverified identity gets a caution note, never a guess. If you went "details only" and the research surfaces someone with a similar name, the report will never assert that person IS your subject — it's framed as "possibly related, unverified" unless your own details substantively match. CoffeeScribe never quietly assumes an identity match it can't back up.
• Deceased subjects. A deceased person's still-live social accounts are labelled estate-run/memorial and are never quoted as if the person wrote them personally — the report's Death & Aftermath section makes this explicit.
• A platform with zero gathered posts is never characterized. If a platform search comes back empty (a quota limit, a dead scrape, whatever the reason), the report says the data wasn't available for that platform rather than inventing a plausible-sounding "active presence" description.

Exporting

The same Copy, Download (.md), Download PDF, and Save to Notebooks (W1.17 — saves the whole report as a note grouped under this Cafe in Notebooks) buttons as every other Cafe report work on a person report. In the PDF, the References section always starts on its own page — again, a shared behaviour across every report type, not person-report-specific. A short AI-generated content notice also appears at the bottom of the report. As of W1.17, a person report is also searchable and citable in Ask Your Cafe, same as every other report type.

---

Cost

Every action shows a live estimate before you spend anything:

Costs are deducted from your token balance like any other Research Mode action; BYOK users' own provider key is used first, with automatic fallback to the platform key (billed to your balance) if your key fails mid-run.

---

Availability

See Research Mode (Cafes) for the rest of what a Cafe can do, and Find the Truth & Claim Evidence Matrix for the sibling feature that checks a specific claim rather than researching a whole person.